How adarshiserit/kannada-sentiment-analysis-model works
KannadaSentiment provides interactive and CLI Kannada emotion inference, plus a separate model-training and evaluation workflow. Training cleans labeled Excel data, creates TF-IDF and MuRIL features, fits classifiers, and evaluates their weighted ensemble; inference combines the saved models’ probabilities. The dashboard supports single-text and batch Excel predictions, while a separate script builds a PDF report from evaluation artifacts. Unsampled orchestrator and dashboard wiring is included only where the README or SOURCE INDEX establishes it.
Main entry points: orchestrator_agent.py Pipeline orchestrator · prediction_agent.py Prediction agent · dashboard.py Streamlit dashboard · generate_report.py PDF report generator
Components 17
Emotion inference
- Streamlit dashboard
app/dashboard.py - CLI predictor
predict.py - Prediction agent
agents/prediction_agent.py
Model training
- Training command
train.py - Pipeline orchestrator
agents/orchestrator_agent.py - Data cleaning
agents/data_agent.py - Feature extraction
agents/feature_agent.py - Classifier training
agents/training_agent.py
Evaluation and reporting
- Ensemble evaluation
agents/evaluation_agent.py - PDF report generator
generate_report.py
Project data and outputs
- Saved models
models - Labeled Excel data
dataset.xlsx - Evaluation artifacts
artifacts - PDF report
artifacts/sentiment_report.pdf
Other components
- Text user
- Excel user
- MuRIL model service
Connections 17
- Text user → Streamlit dashboard: enters text
README.md - Excel user → Streamlit dashboard: uploads workbook
README.md - Streamlit dashboard → Prediction agent: requests predictions
app/dashboard.py - CLI predictor → Prediction agent: requests predictions
predict.py - Prediction agent → Saved models: loads classifiers
agents/prediction_agent.py - Prediction agent → MuRIL model service: loads MuRIL
README.md - Training command → Pipeline orchestrator: starts pipeline
train.py - Labeled Excel data → Data cleaning: provides labeled rows
README.md - Pipeline orchestrator → Data cleaning: loads and cleans
agents/orchestrator_agent.py - Pipeline orchestrator → Feature extraction: extracts features
agents/orchestrator_agent.py - Feature extraction → MuRIL model service: embeds Kannada text
agents/feature_agent.py - Pipeline orchestrator → Classifier training: trains classifiers
README.md - Classifier training → Saved models: saves models
agents/training_agent.py - Pipeline orchestrator → Ensemble evaluation: evaluates ensemble
README.md - Ensemble evaluation → Evaluation artifacts: writes metrics
agents/evaluation_agent.py - PDF report generator → Evaluation artifacts: reads evaluation outputs
generate_report.py - PDF report generator → PDF report: generates report
README.md